This notebook was created to explore contrast-enhancement.

  • I got the idea from the paper, "Segmentation of vessels in angiograms using convolutional neural networks"
    • E. Nasr-Esfahania, et al, 2017 (this was one we discussed months ago)
    • The paper has 2 CNNs, and prior to feeding the CNNs, the images are enhanced with tophat operations

Here is the relevant text from section 2.1, contrast

  • In white top-hat the areas in the image that have high gray scalevalues, i.e. bright areas, are tracked and intensified. In contrast, inblack top-hat the dark regions in the image are highlighted. Henceby applying Eq. (1), after the image is added by white top-hat andthe black top-hat is subtracted from the image, the contrast of theimage is increased.
  • fen = f + fw − fb
  • In multi scale top-hat transform, different scales are selected for structuring element Bi.
  • Using Eqs. (2) and (3) to correspondingblack and white top-hat maps are calculated for each scale.
  • (2) fwi=f− ((fBi) ⊕ Bi)

  • (3)fbi= ((f⊕Bi)  Bi) − f

  • In Eqs. (2) and (3), Bi, f, fwiand fbidenote the structuring ele-ment of the ithscale, input image, the white and black top-hattransforms respectively. Also, ⊕,  are morphological dilation anderosion operators.

  • The dilation and erosion of f(x, y)by Bi(u, v)aredefined by Eqs. (4) and (5).

  • (4) f⊕Bi =maxu,v (f (x−u, y−v) +Bi (u, v))

  • (5) fBi =minu,v (f (x+u, y+v) −Bi (u, v))
In [1]:
import numpy as np
from data_loader import *
import cv2
import skimage
from skimage import morphology
from skimage import filters
import matplotlib.pyplot as plt
%matplotlib inline

C:\Users\Dan\Anaconda3\envs\mcenv\lib\site-packages\dask\config.py:168: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
  data = yaml.load(f.read()) or {}
In [2]:
data, filenames = load_train_test_images(cv=0, stage=1)
        
#
#   Unpack Images from data; all in (n x c x h x w) format
#
train_x_imgs = data["train_x_images"]
train_y_imgs = data["train_y_images"]
test_x_imgs = data["test_x_images"]
test_y_imgs = data["test_y_images"]
train_filenames  = filenames["train_filenames"]
test_filenames = filenames["test_filenames"]
In [3]:
img = train_x_imgs[0]

Subtracting black-tophat from white-tophat to enhance contrast

  • Remember 'B' for Black tophat == Bottom-hat; white-tophat == top-hat
In [4]:
bottom_hat = morphology.black_tophat(np.squeeze(img), selem=morphology.square(20))
tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.square(20))

newimg = img + tophat - bottom_hat

fig = plt.figure(figsize=(16,8));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(newimg), cmap='gray')
Out[4]:
<matplotlib.image.AxesImage at 0x197f8bc5a58>

Using multi-scale tophat transform

  • take max pixel values from tophats with structuring elements of size 3,
In [5]:
elems = [3, 7, 11, 15, 19]
bottom_hats = np.array([morphology.black_tophat(np.squeeze(img), selem=morphology.square(elem)) for elem in elems])
tophats = np.array([morphology.white_tophat(np.squeeze(img), selem=morphology.square(elem)) for elem in elems])

bottom_hat = bottom_hats.max(axis=0)
tophat = tophats.max(axis=0)

newimg = img + tophat - bottom_hat

bottom_hat = morphology.black_tophat(np.squeeze(img), selem=morphology.square(20))
tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.square(20))

newimg = img + tophat - bottom_hat

fig = plt.figure(figsize=(16,8));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(newimg), cmap='gray')
Out[5]:
<matplotlib.image.AxesImage at 0x197f8da1a58>
In [6]:
fig = plt.figure(figsize=(8,40));
plt.title("Original on left, processed on right", fontsize=24);
for i in range(len(tophats)):
    fig.add_subplot(5, 1, i+1)
    plt.imshow(np.squeeze(tophats[i]), cmap='gray')
In [7]:
fig = plt.figure(figsize=(8,8));
t50=morphology.white_tophat(np.squeeze(img), selem=morphology.square(50))
plt.imshow(t50, cmap='gray')
Out[7]:
<matplotlib.image.AxesImage at 0x197f94280f0>
In [8]:
fig = plt.figure(figsize=(8,8));
plt.imshow(cv2.bitwise_not(t50), cmap='gray')
Out[8]:
<matplotlib.image.AxesImage at 0x197f947c6d8>
In [9]:
t50
Out[9]:
array([[34, 30, 33, ..., 29, 20, 20],
       [28, 30, 32, ..., 25, 20, 25],
       [27, 34, 36, ..., 20, 18, 25],
       ...,
       [17, 16, 18, ..., 14, 14, 12],
       [18, 25, 19, ...,  5, 12,  8],
       [14,  9, 10, ...,  3, 14, 12]], dtype=uint8)
In [10]:
cv2.bitwise_not(t50)
Out[10]:
array([[221, 225, 222, ..., 226, 235, 235],
       [227, 225, 223, ..., 230, 235, 230],
       [228, 221, 219, ..., 235, 237, 230],
       ...,
       [238, 239, 237, ..., 241, 241, 243],
       [237, 230, 236, ..., 250, 243, 247],
       [241, 246, 245, ..., 252, 241, 243]], dtype=uint8)

"Blood vessel segmentation in coronary angiogram image" paper

Steps:

  1. The input coronary angiogram image is read.
  2. For preprocessing morphological Top Hat operation is carried out on the input image.
  3. Morphological Closing operation i.e. Dilation and erosion is applied on the image with disk shaped structuring elements.
  4. The image obtained after morphological operation is then thresholded using Otsu’s thresholding .
  5. After thresholding segmented image is obtained.
In [11]:
# Prevent overflow with np.int16
tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(18))
tophat = (np.squeeze(img.astype(np.int16))) + (tophat.astype(np.int16))
pred = np.where(tophat>255,255,tophat).astype(np.uint8)
print("pred max: %.02f, min: %.02f" % (pred.max(), pred.min()))
fig = plt.figure(figsize=(12,6));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(pred), cmap='gray')

pred max: 255.00, min: 39.00
Out[11]:
<matplotlib.image.AxesImage at 0x197f973ba90>
In [12]:
close = morphology.closing(np.squeeze(pred), selem=morphology.disk(18))
close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))
In [13]:
close = morphology.closing(np.squeeze(pred), selem=morphology.disk(18))
close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))
print("close max: %.02f, min: %.02f" % (close.max(), close.min()))

close max: 0.00, min: -165.00
In [14]:
fig = plt.figure(figsize=(12,6));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(close), cmap='gray')
Out[14]:
<matplotlib.image.AxesImage at 0x197f97b4c88>
In [15]:
thresh = filters.threshold_otsu(close)
print("threshold: %.02f"%thresh)
pred=np.where(np.squeeze(close) < thresh, 255, 0)
print("pred max: %.02f, min: %.02f" % (pred.max(), pred.min()))

fig = plt.figure(figsize=(16,8));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(pred), cmap='gray')

threshold: -32.00
pred max: 255.00, min: 0.00
Out[15]:
<matplotlib.image.AxesImage at 0x197f98b7e48>

All together now

In [16]:
allimgs= np.vstack([data["train_x_images"] , data["test_x_images"]])
allfnames = filenames["train_filenames"]+ filenames["test_filenames"]
allimgs.shape, len(allfnames)
Out[16]:
((119, 1, 512, 512), 119)
In [ ]:
# Prevent overflow with np.int16
disksize=22
for i in range(len(allfnames)):
    img = allimgs[i].copy()
    name = allfnames[i]
    #label = train_y_imgs[i].copy()
    
    tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(disksize))
    tophat = (np.squeeze(img.astype(np.int16))) + (tophat.astype(np.int16))
    
    pred=cv2.normalize(tophat,None,0,255,cv2.NORM_MINMAX)
    
    print("pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
    #pred = np.where(tophat>255,255,tophat).astype(np.uint8)
    
    close = morphology.closing(np.squeeze(pred), selem=morphology.disk(disksize))
    close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))
    
    close=cv2.normalize(close,None,0,-255,cv2.NORM_MINMAX)
    print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))
    
    thresh = 1.0 * filters.threshold_otsu(close)
    print("threshold: %.02f"%thresh)
    
    pred=np.where(np.squeeze(close) < thresh, 255, 0)
    
    """
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, processed on right", fontsize=24);
    fig.add_subplot(121)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(pred), cmap='gray')
    """
    
    #plt.imsave("..\\..\\..\\..\\Data\\Vessels\\tophat\\" + name, pred, cmap='gray')

More attempts - trying to follow the instruction on the paper

In [18]:
disksize = 22
img = np.squeeze(train_x_imgs[-1].copy())
label = train_y_imgs[-1].copy()
# Preprocess
tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(disksize))
tophat = img-tophat
print(tophat.max(), tophat.min())
fig = plt.figure(figsize=(12,6));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(tophat), cmap='gray')

159 75
Out[18]:
<matplotlib.image.AxesImage at 0x197fa2d3e48>
In [19]:
tophat = morphology.white_tophat(np.squeeze(cv2.bitwise_not(tophat)), selem=morphology.disk(disksize))
close = morphology.closing(np.squeeze(tophat), selem=morphology.disk(disksize))

fig = plt.figure(figsize=(12,6));
plt.title("Original on left, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(close), cmap='gray')
Out[19]:
<matplotlib.image.AxesImage at 0x197fa34f0f0>
In [ ]:
In [20]:
#pred = np.where(tophat>255,255,tophat).astype(np.uint8)
pred=tophat

close = morphology.closing(np.squeeze(pred), selem=morphology.disk(disksize))
close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))
thresh = filters.threshold_otsu(close)
print("threshold: %.02f"%thresh)
print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))

pred=np.where(np.squeeze(close) < thresh, 255, 0)


fig = plt.figure(figsize=(18,6));
plt.title("Original on left, label in middle, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(132)
plt.imshow(np.squeeze(label), cmap='gray')
fig.add_subplot(133)
plt.imshow(np.squeeze(pred), cmap='gray')

threshold: -3.00
close max: 0.00, min: -13.00, range: 13.00
Out[20]:
<matplotlib.image.AxesImage at 0x197f92b9c88>

More attempts - try to run process twice on image

In [21]:
img = allimgs[0].copy()
name = allfnames[0]
#label = train_y_imgs[i].copy()

tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(disksize))
tophat = (np.squeeze(img.astype(np.int16))) + (tophat.astype(np.int16))

pred=cv2.normalize(tophat,None,0,255,cv2.NORM_MINMAX)

print("pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
#pred = np.where(tophat>255,255,tophat).astype(np.uint8)

close = morphology.closing(np.squeeze(pred), selem=morphology.disk(disksize))
close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))

close=cv2.normalize(close,None,0,-255,cv2.NORM_MINMAX)
print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))

thresh = 1.0 * filters.threshold_otsu(close)
print("threshold: %.02f"%thresh)

pred=np.where(np.squeeze(close) < thresh, 255, 0)

fig = plt.figure(figsize=(18,6));
plt.title("Original on left, label in middle, processed on right", fontsize=24);
fig.add_subplot(131)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(132)
plt.imshow(np.squeeze(label), cmap='gray')
fig.add_subplot(133)
plt.imshow(np.squeeze(pred), cmap='gray')

pred max: 255.00, min: 0.00, range: 255.00
close max: 0.00, min: -255.00, range: 255.00
threshold: -56.00
Out[21]:
<matplotlib.image.AxesImage at 0x197f927fcc0>
In [22]:
fig = plt.figure(figsize=(6,6));
saveclose=cv2.normalize(close,None,0,255,cv2.NORM_MINMAX).astype(np.uint8)
print(saveclose.max(), saveclose.min())
plt.imshow(close, cmap='gray')

255 0
Out[22]:
<matplotlib.image.AxesImage at 0x197fa0cd8d0>
In [23]:
test = skimage.exposure.equalize_adapthist(saveclose)
fig = plt.figure(figsize=(6,6));
test=cv2.normalize(test,None,0,255,cv2.NORM_MINMAX).astype(np.uint8)
print(test.max(), test.min())
plt.imshow(test, cmap='gray')

255 0
Out[23]:
<matplotlib.image.AxesImage at 0x197fa126940>
In [24]:
thresh = 1.0 * filters.threshold_otsu(test)
print("threshold: %.02f"%thresh)

pred=np.where(np.squeeze(test) < thresh, 255, 0)

fig = plt.figure(figsize=(18,6));
plt.title("Original on left, label in middle, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(pred), cmap='gray')

threshold: 176.00
Out[24]:
<matplotlib.image.AxesImage at 0x197f9d57b70>
In [25]:
test2 = skimage.exposure.equalize_adapthist(test)
fig = plt.figure(figsize=(6,6));
test2=cv2.normalize(test2,None,0,255,cv2.NORM_MINMAX).astype(np.uint8)
print(test2.max(), test2.min())
plt.imshow(test2, cmap='gray')

255 0
Out[25]:
<matplotlib.image.AxesImage at 0x197f9dfe160>
In [26]:
thresh = 1.0 * filters.threshold_otsu(test2)
print("threshold: %.02f"%thresh)

pred=np.where(np.squeeze(test2) < thresh, 255, 0)

fig = plt.figure(figsize=(18,6));
plt.title("Original on left, label in middle, processed on right", fontsize=24);
fig.add_subplot(121)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(122)
plt.imshow(np.squeeze(pred), cmap='gray')

threshold: 144.00
Out[26]:
<matplotlib.image.AxesImage at 0x197fc658c88>
In [27]:
test3 = skimage.exposure.equalize_adapthist(test2)
fig = plt.figure(figsize=(6,6));
test3=cv2.normalize(test3,None,0,255,cv2.NORM_MINMAX).astype(np.uint8)
print(test3.max(), test3.min())
plt.imshow(test3, cmap='gray')

255 0
Out[27]:
<matplotlib.image.AxesImage at 0x197fc6fada0>
In [28]:
tophat = morphology.white_tophat(np.squeeze(saveclose), selem=morphology.disk(disksize))
tophat = (np.squeeze(saveclose.astype(np.int16))) + (tophat.astype(np.int16))

pred=cv2.normalize(tophat,None,0,255,cv2.NORM_MINMAX)

print("pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
#pred = np.where(tophat>255,255,tophat).astype(np.uint8)

close = morphology.closing(np.squeeze(pred), selem=morphology.disk(disksize))
close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))

close=cv2.normalize(close,None,0,-255,cv2.NORM_MINMAX)
print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))

thresh = 1.0 * filters.threshold_otsu(close)
print("threshold: %.02f"%thresh)

pred=np.where(np.squeeze(close) < thresh, 255, 0)

fig = plt.figure(figsize=(18,6));
plt.title("Original on left, label in middle, processed on right", fontsize=24);
fig.add_subplot(131)
plt.imshow(np.squeeze(img), cmap='gray')
fig.add_subplot(132)
plt.imshow(np.squeeze(label), cmap='gray')
fig.add_subplot(133)
plt.imshow(np.squeeze(pred), cmap='gray')

pred max: 255.00, min: 0.00, range: 255.00
close max: 0.00, min: -255.00, range: 255.00
threshold: -57.00
Out[28]:
<matplotlib.image.AxesImage at 0x197fcf8c550>
In [ ]:
In [ ]:
In [29]:
allimgs= np.vstack([data["train_x_images"] , data["test_x_images"]])
allfnames = filenames["train_filenames"]+ filenames["test_filenames"]
allimgs.shape, len(allfnames)
Out[29]:
((119, 1, 512, 512), 119)
In [30]:
# Prevent overflow with np.int16
disksize=22
for i in range(1):
    img = allimgs[i].copy()
    name = allfnames[i]
    label = train_y_imgs[i].copy()
    
    tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(disksize))
    tophat = (np.squeeze(img.astype(np.int16))) + (tophat.astype(np.int16))
    
    pred=cv2.normalize(tophat,None,0,255,cv2.NORM_MINMAX)
    
    print("pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
    #pred = np.where(tophat>255,255,tophat).astype(np.uint8)
    
    close = morphology.black_tophat(np.squeeze(pred), selem=morphology.disk(disksize))
    #close = (np.squeeze(pred.astype(np.int16))) - (close.astype(np.int16))
    
    close=cv2.normalize(close,None,0,255,cv2.NORM_MINMAX)
    print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))
    
    thresh = 1.0 * filters.threshold_otsu(close)
    print("threshold: %.02f"%thresh)
    
    pred=np.where(np.squeeze(close) > thresh, 255, 0)
    
    
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, processed on right", fontsize=24);
    fig.add_subplot(131)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(pred), cmap='gray')
    

pred max: 255.00, min: 0.00, range: 255.00
close max: 255.00, min: 0.00, range: 255.00
threshold: 54.00
In [31]:
# Prevent overflow with np.int16
disksize=22
for i in range(1):
    img = allimgs[i].copy()
    name = allfnames[i]
    label = train_y_imgs[i].copy()
    
    tophat = morphology.white_tophat(np.squeeze(img), selem=morphology.disk(disksize))
    tophat = (np.squeeze(img.astype(np.int16))) + (tophat.astype(np.int16))
    
    preprocess=cv2.normalize(tophat,None,0,255,cv2.NORM_MINMAX)
    
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, after tophat on right", fontsize=24);
    fig.add_subplot(131)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(preprocess), cmap='gray')
    
    print("pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
    #pred = np.where(tophat>255,255,tophat).astype(np.uint8)
    
    # Reverse and tophat again
    #pred = cv2.bitwise_not(preprocess.copy())
    pred = 255 - preprocess
    tophat = morphology.white_tophat(np.squeeze(pred), selem=morphology.disk(disksize))
    tophat = (np.squeeze(pred.astype(np.int16))) + (tophat.astype(np.int16)) - preprocess
    
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, new tophat on right", fontsize=24);
    fig.add_subplot(131)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(pred), cmap='gray')
    
    print("new tophat pred max: %.02f, min: %.02f, range: %.02f" % (pred.max(), pred.min(), pred.max()-pred.min()))
    #pred = np.where(tophat>255,255,tophat).astype(np.uint8)
    
    close = morphology.closing(np.squeeze(pred), selem=morphology.disk(disksize))
    close = (np.squeeze(pred.astype(np.int16))) + (close.astype(np.int16))
    
    close=cv2.normalize(close,None,0,255,cv2.NORM_MINMAX)
    print("close max: %.02f, min: %.02f, range: %.02f" % (close.max(), close.min(), close.max()-close.min()))
    
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, after close on right", fontsize=24);
    fig.add_subplot(131)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(close), cmap='gray')
    
    thresh = 1.0 * filters.threshold_otsu(close)
    print("threshold: %.02f"%thresh)
    print(allfnames[i])
    
    pred=np.where(np.squeeze(close) < thresh, 255, 0)
    
    
    fig = plt.figure(figsize=(18,6));
    plt.title("Original on left, label in middle, processed on right", fontsize=24);
    fig.add_subplot(131)
    plt.imshow(np.squeeze(img), cmap='gray')
    fig.add_subplot(132)
    plt.imshow(np.squeeze(label), cmap='gray')
    fig.add_subplot(133)
    plt.imshow(np.squeeze(pred), cmap='gray')
    

pred max: 255.00, min: 0.00, range: 255.00
new tophat pred max: 255.00, min: 0.00, range: 255.00
close max: 255.00, min: 0.00, range: 255.00
threshold: 133.00
23_LCA_RAO
In [32]:
#fig = plt.figure(figsize=(18,6));
##plt.title("pred on left, denoise bilateral center, tf chambolle on right", fontsize=24);
#fig.add_subplot(131)
#plt.imshow(np.squeeze(pred), cmap='gray')
#fig.add_subplot(132)
#plt.imshow(np.squeeze(skimage.restoration.denoise_bilateral(pred, sigma_color=0.05, sigma_spatial=15, multichannel=False)), cmap='gray')
#fig.add_subplot(133)
#plt.imshow(np.squeeze(skimage.restoration.denoise_tv_chambolle(pred, weight=0.2, multichannel=False)), cmap='gray')
In [33]:
fig = plt.figure(figsize=(6,6));
plt.imshow(cv2.bitwise_not(np.squeeze(close)), cmap='gray')
Out[33]:
<matplotlib.image.AxesImage at 0x19782433550>
In [34]:
train_filenames[0]
Out[34]:
'23_LCA_RAO'
In [ ]:
In [ ]: